forked from cloud/ovgate
A userspace VPN gateway: builds an OpenVPN tunnel to a VPNGate node with
the OpenVPN 3 core, terminates it in-process with lwIP, and serves SOCKS5
(RFC 1928/1929, CONNECT and UDP ASSOCIATE) over it. No root, no tun
device, no routing table changes.
Layout follows the module boundaries in docs/ARCHITECTURE.md:
vpngate/ directory fetch + CSV parse (lines run to ~13.5 KB, so the
parser streams rather than splitting on newlines)
selector/ two-phase pick: cheap prior over the whole list, then real
TCP handshake timing of the top K
ovpn/ openvpn3 driven through TunBuilder, packets over a socketpair
netstack/ lwIP: the TCP/IP stack that makes "no root" possible
egress/ the swappable way out, and make-before-break switching
socks5/ the front door
health/ per-window scoring, and the decision to move
app/ wiring, admin HTTP, signals
docs/FEASIBILITY.md is the analysis this was built from, including the
one requirement that is not physically possible -- carrying established
TCP connections across a node switch -- and what is done instead
(zero-progress redial, UDP re-homing, grace-period drain).
Tests: 155 without the tunnel egress, 172 with it. The seam is the egress
factory; selection, scoring, history and probing all run for real.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
154 lines
5.7 KiB
C++
154 lines
5.7 KiB
C++
// The way out. Everything above this line -- SOCKS5, health, admin -- programs
|
|
// against this interface and nothing else.
|
|
//
|
|
// ---------------------------------------------------------------------------
|
|
// Why the whole design hangs on this being a shared_ptr
|
|
// ---------------------------------------------------------------------------
|
|
// A SOCKS5 session takes a reference at accept time and holds it until it ends.
|
|
// That single rule buys three things that would otherwise each need their own
|
|
// machinery:
|
|
//
|
|
// * A session can never be re-homed underneath itself. The egress it started
|
|
// on is the egress it finishes on, so there is no window where a half-sent
|
|
// request goes out one tunnel and its continuation out another.
|
|
// * Draining needs no session table. The reference count *is* the number of
|
|
// sessions still using this egress; when it hits zero the destructor closes
|
|
// the tunnel. Nothing to iterate, nothing to leak.
|
|
// * The switch is a pointer swap. Promote = store a new shared_ptr in the
|
|
// manager; every acquire() after that gets the new one, every session
|
|
// before it is untouched.
|
|
//
|
|
// docs/ARCHITECTURE.md §5.2.
|
|
//
|
|
// ---------------------------------------------------------------------------
|
|
// Threading
|
|
// ---------------------------------------------------------------------------
|
|
// Every method is safe to call from any thread. Each async method takes the
|
|
// executor its handler must run on, and so does everything the handler hands
|
|
// back: a stream created with a session's strand delivers all of its own
|
|
// completions on that strand too.
|
|
//
|
|
// Passing the executor explicitly rather than fixing it at construction is what
|
|
// lets one egress serve a thousand sessions on four io threads while each
|
|
// session still sees a single-threaded world. Implementations that need
|
|
// internal serialization -- the tunnel one does, because lwIP does -- arrange
|
|
// that privately and never leak it to the caller.
|
|
#pragma once
|
|
|
|
#include <asio.hpp>
|
|
|
|
#include <cstdint>
|
|
#include <functional>
|
|
#include <memory>
|
|
#include <string>
|
|
#include <vector>
|
|
|
|
#include "common/config.h"
|
|
#include "common/endpoint.h"
|
|
#include "netstack/stream.h"
|
|
|
|
namespace ovg::egress {
|
|
|
|
using netstack::TcpStreamPtr;
|
|
using netstack::UdpSocketPtr;
|
|
|
|
enum class EgressState {
|
|
Idle, // constructed, start() not called
|
|
Connecting, // bringing the tunnel up; no traffic yet
|
|
Ready, // usable
|
|
Draining, // still serving existing sessions, refusing new ones
|
|
Down, // unusable; `detail()` says why
|
|
};
|
|
const char *egress_state_name(EgressState s);
|
|
|
|
struct EgressStats {
|
|
// Identity.
|
|
std::string node_id;
|
|
std::string node_country;
|
|
std::string server_ip;
|
|
std::string local_address; // our address inside the tunnel
|
|
std::string proto; // "udp" / "tcp" / "direct"
|
|
|
|
// Age and use.
|
|
int64_t uptime_ms = 0;
|
|
int64_t sessions = 0; // shared_ptr use_count minus the manager's own refs
|
|
|
|
// Traffic, as seen at the tunnel. Zero for the direct egress, which has no
|
|
// single aggregate to report.
|
|
uint64_t tun_bytes_in = 0;
|
|
uint64_t tun_bytes_out = 0;
|
|
uint64_t transport_bytes_in = 0;
|
|
uint64_t transport_bytes_out = 0;
|
|
int last_packet_received_ms = -1; // -1 = never
|
|
|
|
// Connection outcomes, since construction.
|
|
uint64_t tcp_opened = 0;
|
|
uint64_t tcp_failed = 0;
|
|
int64_t tcp_active = 0;
|
|
int64_t udp_active = 0;
|
|
|
|
// Packet-level health, from the netstack. All zero for direct.
|
|
uint64_t rx_packets = 0;
|
|
uint64_t tx_packets = 0;
|
|
uint64_t tx_dropped = 0;
|
|
uint64_t rx_malformed = 0;
|
|
uint64_t rx_dropped = 0;
|
|
|
|
double connect_failure_rate() const {
|
|
const uint64_t total = tcp_opened + tcp_failed;
|
|
return total == 0 ? 0.0 : static_cast<double>(tcp_failed) / total;
|
|
}
|
|
};
|
|
|
|
class Egress {
|
|
public:
|
|
using ConnectHandler =
|
|
std::function<void(const std::error_code &, TcpStreamPtr)>;
|
|
using UdpBindHandler =
|
|
std::function<void(const std::error_code &, UdpSocketPtr)>;
|
|
using ResolveHandler =
|
|
std::function<void(const std::error_code &, std::vector<IpAddress>)>;
|
|
|
|
virtual ~Egress() = default;
|
|
|
|
// `target` may be a domain: resolving it here rather than at the caller is
|
|
// what keeps DNS inside the tunnel. A literal address skips the lookup.
|
|
virtual void async_connect_tcp(const asio::any_io_executor &ex,
|
|
const Endpoint &target, Millis timeout,
|
|
ConnectHandler h) = 0;
|
|
|
|
// An outbound datagram socket on this egress, for SOCKS5 UDP ASSOCIATE.
|
|
virtual void async_bind_udp(const asio::any_io_executor &ex,
|
|
UdpBindHandler h) = 0;
|
|
|
|
// Exposed separately from connect because SOCKS5 needs the resolved address
|
|
// for the BND field in the reply, and the health monitor times a lookup as
|
|
// its in-tunnel RTT probe.
|
|
virtual void async_resolve(const asio::any_io_executor &ex,
|
|
const std::string &host, ResolveHandler h) = 0;
|
|
|
|
virtual EgressState state() const = 0;
|
|
virtual EgressStats stats() const = 0;
|
|
|
|
// Human-readable reason for Down, or an empty string.
|
|
virtual std::string detail() const = 0;
|
|
|
|
// Stops accepting new work. Existing streams keep running: this is the state
|
|
// an egress sits in between being replaced and its last session ending.
|
|
// Idempotent.
|
|
virtual void begin_drain() = 0;
|
|
|
|
// Tears everything down now, aborting live streams with Error::EgressGone.
|
|
// `on_done` runs on the egress's executor once the tunnel is gone.
|
|
virtual void shutdown(std::function<void()> on_done = {}) = 0;
|
|
|
|
// Stable label for logs: node id for a tunnel, "direct" otherwise.
|
|
virtual const std::string &label() const = 0;
|
|
|
|
bool usable() const { return state() == EgressState::Ready; }
|
|
};
|
|
|
|
using EgressPtr = std::shared_ptr<Egress>;
|
|
|
|
} // namespace ovg::egress
|